The Role of the Securitization Process in the Expansion of Subprime Credit

Transcription

1 The Role of the Securitization Process in the Expansion of Subprime Credit Taylor D. Nadauld * Doctoral Candidate Department of Finance The Ohio State University Nadauld_1@fisher.osu.edu Shane M. Sherlund* Division of Research and Statistics Board of Governors of the Federal Reserve System First Draft: May 2008 This Draft: August 2008 Abstract We analyze the structure and attributes of subprime mortgage-backed securitization deals originated between Our data set allows us to map loan-level data for over 6.5 million subprime loans to the securitization deals into which the loans were sold. We document the relationship between the structure of the securitization deal and the attributes of the underlying mortgage collateral, including housing market conditions at the time of deal origination. We find evidence that deals with higher levels of housing market diversification have a larger portion of the deal rated investment grade. Consistent with our primary hypothesis we find that deals comprised of loans concentrated in areas with high rates of home price appreciation also have a larger portion of the deal rated investment grade. We believe these results highlight how the structure of securitization deals could impact the supply of credit being afforded the mortgage origination market. Deal structure matters because the economics of the structuring process create incentives for deal arrangers to purchase loans that will provide the cheapest funding for the deal. Journal of Economic Literature classification numbers: G21, G24. Keywords: Securitization, subprime mortgages, financial intermediation. * We are grateful to Adam Ashcraft, Mike Andersen, Josh Coval, Karl Diether, Andrew Karolyi, Rose Liao, Anil Makhija, Anthony Sanders, Rene Stulz, Jerome Taillard, Michael Weisbach, Scott Yonker, and participants in the Finance seminar at The Ohio State University for helpful comments and suggestions.

2 The default rate on residential subprime mortgage loans sold into securitization deals has nearly tripled since Rising default rates have resulted in substantial losses for investors holding investment-grade mortgage-backed securities (MBS) as well as holders of collateralized-debt obligations (CDOs) which employ MBS as collateral. The rise in defaults has taken its toll on credit markets in general. As of July 17, 2008, Bloomberg reported that banks and brokers have taken more than $435 billion in writedowns due to the poor performance of MBS, CDOs and other forms of leveraged loans. The observed phenomenon begs an obvious question: Why has the default rate of mortgage loans increased so dramatically? 2 A recent literature points to an increase in the supply of credit made to subprime borrowers in mortgage origination markets, allowing borrowers of marginal credit quality to obtain loans (Sufi and Mian 2008, Dell'Ariccia, Igan, and Laeven 2008). The cause of the increase in credit supply (or decrease in credit standards), it is argued, is increased securitization activity. If true, this suggests an important additional question: What role did the structure of mortgage-backed securitizations play in increased subprime defaults? Our study seeks to answer this question by examining the securitization process and structure of over 1,250 subprime mortgaged-backed securitization deals originated between 1997 and 2007, the attributes of the 6.7 million individual loans that comprise the securitization deals, and the macroeconomic conditions that existed in the areas where the loans were originated. In ascribing the increase in the extension of subprime credit to heightened securitization activity, the literature has thus far relied on binary classifications of securitization; loans were either securitized or not. Our study highlights how the structure and ratings of subprime residential mortgage-backed securitization deals affect the economic incentives driving the loan purchase decisions of investment banks in the secondary mortgage market. 3 We focus on how the credit and housing market 1 According to data from LoanPerformance, subprime loans included in securitization deals that were originated in 2002 had a 4.01% default rate by the end of By comparison, loans originated and sold into securitization deals in 2006 had a default rate of 13.7% by December One immediate answer is the decline in home prices. However, to the extent that subprime loans, particularly subprime loans with adjustable rates are the ones most affected by price declines, the relevant question is why such loans were originated. 3 Figure 1a presents a schematic of the impact of securitization on the credit threshold in the mortgage origination market. Figure 1b documents the feedback effect that deal structure could have on the credit threshold. Loans that deliver favorable ratings in the securitization process will be originated in the primary market. 2

3 characteristics of the loans included in mortgage-backed securitization deals affect the terms on which a deal is arranged. Our primary finding is that deals concentrated in areas with high rates of house price appreciation have a larger percentage of the deal rated investment grade. Deals with a larger portion of the principal rated investment grade can fund the purchase of the underlying loan collateral at a lower cost because they can issue bonds with lower coupon payments (higher prices). The implication is straightforward. If higher rates of house price appreciation can lower the cost of funding a deal, the secondary market will demand loans in rapidly appreciating housing markets, and mortgage originators will rationally increase the supply. Our study is not an investigation of whether the credit ratings assigned to subprime securitization deals were correct. Rather, we take the ratings process as given and seek to understand the economic implications of the securitization and rating process. Our study is motivated primarily by a theory offered by Ashcraft and Schuermann (2008) which outlines how expected rates of house price appreciation can affect securitization deal structure, and eventually the supply of credit afforded the origination market. To the extent that mortgage pools concentrated in areas of high house price appreciation receive more favorable credit ratings (all else being equal), deal arrangers could rationally purchase mortgages of a lower marginal credit quality that are concentrated in areas of high price appreciation and still obtain the investment-grade ratings required to profitably market a securitization deal. We test the implications of this hypothesis in three ways. First, we test whether deals with loans concentrated in areas with high rates of house price appreciation are indeed able to get a larger portion of the securitization rated investment grade, and whether this translates into a lower cost of funds for the deal. Second, we test whether pools of loans concentrated in areas of high house price appreciation are of a lower average credit quality. Finally, a potential consequence of purchasing loans of marginal credit quality masked by high rates of house price appreciation is the potential for higher default rates when rates of house price appreciation slow down (or even turn negative). Accordingly, we analyze whether realized rates of house price appreciation have a significant association with deal-level default rates. 3

4 Aside from the considerable interest subprime loans are receiving from the media and investment community, we ask why the subprime experience should matter outside of an isolated episode. Prior literature attributes the profitable practice of pooling and tranching cash flows to the presence of asymmetric information (DeMarzo (2005)), or incomplete markets (Gaur, Seshadri, and Subrahmanyam (2003)). After all, in a world of perfect capital markets, why should the repackaging of cash flows be a profitable enterprise? Two recent studies, relying on the assumption that investors purchase bonds based on credit ratings, explain the proliferation of securitization activity to the potential for deal arrangers to deliver the cheapest possible set of assets that can obtain a high quality credit rating. Coval, Jurek, and Stafford (2007) conclude that the growth of the credit tranche market can potentially be explained as an endogenous, institutional response to an arbitrage opportunity in the credit markets. In particular, Coval et al. demonstrate that because credit ratings do not account for the state in which defaults occur, naïve prices based solely on ratings will not account for systematic, priced risk factors. Brennan, Hein, and Poon (2008) also attribute the existence of pooling and tranching to potential ratings arbitrage. Our sample of subprime securitizations may provide some limited, indirect evidence consistent with a theory of ratings arbitrage. To the extent that investors purchase MBS based solely on credit ratings, the key assumption in models of ratings arbitrage, investment banks are incented to deliver the cheapest portfolio of loans (which would be loans of lower credit quality) that can still obtain an investment-grade rating. Because we don t have wholesale prices of the underlying mortgage loan collateral, we cannot directly test this hypothesis. 4 However, our analysis of the association between rates of house price appreciation and pool credit quality could be viewed as an indirect test of deal arrangers ability to arbitrage the ratings process by purchasing cheaper loans (loans of a marginal credit quality) that are able to deliver marketable credit ratings on account of high rates of house price appreciation. Throughout our empirical analysis, we essentially make the assumption that arrangers of securitization deals (usually, investment banks) seek to maximize the portion 4 One concern in an MBS setting would be that, although loans of a lesser credit quality are likely priced at a discount, the wholesale loan market could conceivably command a premium for loans concentrated in areas of high expected appreciation. 4

5 of each securitization deal that receives investment-grade ratings. 5 All else being equal, maximizing the weighted average of ratings in the deal lowers the cost of funding for the deal. To test our hypotheses, we calculate the total percentage of deal principal that received a AAA rating, and the percent receiving an investment-grade credit rating (BBB+ or higher (or the Moody s equivalent)). We then evaluate the total percentage of a deal rated AAA or investment grade against deal-level measures of house price appreciation. In so doing, we develop proxies to control for the potential default correlation in the underlying pool of collateral. In addition to default correlation, which impacts the shape of the default distribution, we control for other important loan attributes that predict expected loss in a pool of mortgage loans, such as FICO scores, loan-to-value ratios, loan type, loan purpose, and macroeconomic factors. Our sample also allows us to provide a description of the subprime securitization landscape. We document deal size and structure, deal frequency, and the changing characteristics of mortgage loans included in the deals through time and in the cross section. We find the strongest support economically for our hypothesis related to rates of house price appreciation. On average, a 5% increase in the one-year lagged rate of house price appreciation is associated with a 1%-2% increase in the percentage of a deal rated investment grade (one standard deviation in the percent rated investment grade is 4%). The result is strongest in the year As expected, we also find that deals with lower levels of geographic concentration are structured on better terms, though the results are not large economically. Decreasing the geographic concentration of a deal by 5% (one standard deviation) increases the total proportion of the deal rated AAA by 0.4% (one standard deviation equals 6%). Our limited, indirect test of a ratings arbitrage hypothesis, which employs a simple two-way sorting technique, finds support for the hypothesis that high rates of house price appreciation allow banks to purchase loans of a lower marginal credit quality while maintaining favorable credit ratings. The results suggest that portfolios of subprime loans with similar credit ratings but disparate rates of house price appreciation vary widely in the percent of loans with an adjustable rate. Lastly, in an 5 One potential objection to this assumption would be theories that rely on market incompleteness as motivation for securitization (Gaur, Seshardri, and Subrahmanyam (2003)). A market incompleteness theory argues that firms engineer securities from the securitization process that complete the market. If completing markets is the primary motive for securitization, it is not obvious that a deal arranger s primary objective is to obtain the best possible ratings for a deal. 5

6 effort to highlight the implications of our results, we identify strong associations between loan default rates and contemporaneous changes in housing prices. Recent literature on the subprime crisis addresses a number of questions relevant to this paper. Sufi and Mian (2008) argue that a shift in the supply of credit made to subprime borrowers caused an increase in house prices and subsequent default rates. They attribute the increase in credit supply to the existence of securitization. Our results suggest that the relationship between house price appreciation and credit supply is at the very least simultaneous. 6 While an increase in the supply of credit may have increased house prices, it appears that rates of house price appreciation themselves can affect the supply of credit. Our results also build upon Sufi and Mian (2008) in that we identify how the securitization process may have caused an unexpected increase in credit supply. Our results provide evidence of the incentives for investment banks to purchase a higher proportion of loans in areas of high price appreciation, a process that could work as a mechanism rationing credit to the origination market. Mayer and Pence (2008) provide strong evidence in support of this premise; mortgage origination rates, a proxy for the supply of credit in the primary mortgage market, exhibit a positive, robust relationship with lagged rates of house price appreciation. Dell'Ariccia, Igan, and Laeven (2008) demonstrate that lending standards declined in areas of high home price appreciation and attribute the decline in lending standards to increased competition among lenders. Keys, Mukherjee, Seru, and Vig (2008) show that securitized loans with a credit score slightly above the traditional subprime threshold (FICO 620) were 20% more likely to default then securitized loans slightly below the subprime threshold. The result is interpreted as evidence that the prospect of selling loans to secondary markets reduces lenders' incentives to carefully screen borrowers. Demyanyk and Van Hemert (2008) find that credit quality was inexplicably low during , even after controlling for house price appreciation. Gabriel and Rosenthal (2007) present and estimate a model explaining how active secondary markets can increase the supply of credit to primary borrowers. In sum, the previous literature has focused almost exclusively on the loan origination market and binary classifications for 6 Sufi and Mian use a measure of pent up demand as an instrument for credit supply. 6

7 securitized loans. This paper, to the best of our knowledge, is unique in exploring the importance of the securitization process in the subprime crisis. Our paper is organized as follows. Section 2 documents the institutional features of the securitization market. Section 3 provides a discussion of the theories of securitization and motivates our hypotheses. Section 4 describes the data. In section 5 we describe our empirical strategy and discuss results. Section 6 concludes. Section 2: Institutional Features of the Securitization Market Section 2a: Deal Structure In order to understand how the securitization process can impact mortgage markets, we provide a brief outline of some of the key institutional features of the subprime securitization structure. Though no strict definition of a subprime mortgage exists, the term usually refers to a mortgage loan with poor credit quality, excessive leverage, or no income documentation. Borrowers who "state" a monthly income without documentation to verify the income can also be considered subprime. 7 High loan-tovalue or debt-to-income ratios are also typical of subprime borrowers. Until the late 1990s, subprime loans represented a very small portion of total residential mortgage originations. Pools of subprime loans are originated by retail banks or mortgage brokers and subsequently sold to private financial intermediaries, who are frequently Wall Street firms or their subsidiaries, and who are referred to as deal arrangers or deal underwriters. 8 The pool of loans is then placed into a bankruptcy remote trust, which is a separate legal entity and which owns the rights to each mortgage. Servicing of the mortgages is outsourced to a loan servicer. The pool of loans is separated into different "tranches" against which bonds are issued and sold to investors. Mortgage payments from the pool of loans are passed through to the bond holders and are the source of the bond's coupon payment. The type of bonds issued against securitized mortgages varies 7 Borrowers who state a monthly income can also fall into a category of loans called Alt-A. Again, though no strict definition exists, Alt-A loans generally have higher credit scores than subprime borrowers but lack income documentation. 8 In Appendix A3 we provide a list of firms who acted as deal underwriters in our sample of securitization deals. 7

8 substantially. Bond coupon payments can be fixed or variable. Some bonds are issued as interest only, so that the bondholder receives only the interest from an underlying mortgage pool, while others are issued as principal only. Bonds referred to as the equity tranche generally do not receive any principal. 9 Individual loans are not assigned to specific tranches. Rather, tranches are organized in a seniority structure that assigns a priority payment scheme to payment streams emanating from the underlying loans. The prioritized payment of principal and interest varies by deal. Typically, the principal from loans that pre-pay (refinance or sell) before their stated maturities flows first to holders of senior tranches, while defaults first reduce the principal of the most junior tranches until their principal is exhausted. Holders of junior tranches are subject to default risk, or the risk that the principal balance of mortgages from which coupon payments flow will be eroded. Figure 2 displays a simple diagram of a sample securitization structure from our data. The deal, originated by Goldman Sachs in February 2006 had a total deal principal of $714.2 million. The figure reports the size of each tranche, its original credit rating and the rate of the first scheduled coupon payment. For a more detailed discussion of the institutional features of the subprime securitization market, we refer the interest reader to Ashcraft and Schuermann (2008). Section 2b: Subordination, Excess Spread, and Other Forms of Credit Enhancement Pre-specified cash flow rules are designed to ensure that bonds with investmentgrade ratings receive the promised coupon payments with a very high probability, ex ante. In order to ensure that holders of investment-grade bonds receive the promised payments, deals receive credit support against the potential for mortgage defaults. The credit support works to protect senior tranches against the loss of coupon payments 9 Equity tranches can receive payments at the beginning of their life. Deals with large margins between the underlying mortgage rate and coupon payments can pass the excess interest payments onto holders of the equity tranche. These payments only occur when every other tranche is receiving its full coupon payment. In practice, this only occurs in the infancy of a deal. When default rates increase, the excess margin is required to compensate more senior tranches. We discuss this concept of excess spread in much greater detail in section 2b. 8

9 stemming from default. The two most prevalent forms of credit support are subordination and excess spread. 10 For any given tranche, subordination is the sum of the amount of principal that exists in any junior tranches. For example, if all tranches with an S&P credit rating of AAA represented 80% of the total principal in a deal, the AAA tranches are said to benefit from 20% subordination. Subordination for investment-grade tranches, which are those with an S&P credit rating of BBB+ (or Moody s equivalent) or higher, is defined in the same way. In our sample, the median proportion of deal principal rated AAA is 79.9%. The median proportion of principal rated investment grade is 94.5%. Portions of the securitization structure not rated investment grade are generally made up of one or two very small non-investment grade bonds that pay high coupons, and a tranche referred to as over-collateralization. The over-collateralization tranche does not pay a coupon and exists solely to provide credit protection to more senior tranches. Loans that default first will destroy the principal balance of the over-collateralization piece before touching any tranche more senior. Only after the over-collateralization principal has been fully exhausted will defaults accrue to the next most junior tranche. Thus, senior tranches benefit from thick junior tranches, and in this way, subordination acts as a form of credit protection. Excess spread is the second form of credit protection that exists to insure senior tranches against mortgage default. Excess spread is defined as the difference between the payments coming into the securitization structure from the underlying mortgage collateral and the rate being paid to coupon holders. Excess spread is calculated net of fees paid to mortgage servicers and other intermediaries, such as interest-rate swap counterparties. As Ashcraft and Schuermann (2008) explain, [excess spread] is the first line of defense for investors against credit losses, as no amount of principal of any tranche is reduced by any amount until credit losses reduce excess spread to a negative number. In this way, higher levels of excess spread provide more credit protection to holders of senior tranches. 10 The term credit support is used interchangeably in the literature with the term credit enhancement or credit protection. We will also use the terms interchangeably. 9

10 Deals benefit from other forms of credit enhancement such as shifting interest, performance triggers, and interest rate swaps. Shifting interest requires that all pre-paid principal be applied only to senior tranches for a pre-specified period (typically the first 36 months). The practice of shifting interest serves to increase the subordination of senior tranches because pre-payments reduce their principal balance, leaving their principal as a smaller percentage of the total deal principal. Performance triggers exist to ensure that pre-paid principal is not released to any class until the deal passes pre-specified performance tests. 11 Thus, if a deal is not performing well, the priority rules can be shifted to ensure senior tranches receive proper credit support. Finally, deals with floating coupon payments manage the risk that coupon payments to bondholders might rise faster than rates on the underlying mortgages by means of interest rate swaps. 12 Section 3: Hypothesis Development One of the key issues associated with the increase in subprime defaults is the observed increase in the extension of mortgage loans to borrowers of marginal credit quality (Dell'Ariccia, Igan, and Laeven, 2008). A critical question is why this occurred. The literature has suggested the existence of securitization as one explanation (Sufi and Mian (2008), Keys et al. (2008)). 13 The purpose of our paper is to clarify how the specific structure of securitization deals and the ratings process could influence the types of loans investment banks would purchase in the secondary mortgage market for the purpose of issuing securitized bonds. By identifying the loan attributes that maximize the gains for investment banks we can better understand why origination markets would have rationally originated loans of marginal credit quality, particularly in areas with high rates of house price appreciation. 11 We do not have data on which deals in our sample benefit from the existence of performance triggers. 12 A trust making floating coupon payments backed by a pool of fixed-rate mortgages could hedge the interest rate risk by entering into a swap agreement to pay fixed rates to a counterparty in exchange for variable interest payments. 13 As predicted by Gorton and Pennacchi (1993) and Diamond (1984), originators of loans have little incentive to carefully screen or monitor borrowers if they do not bear the risk of the loans. This isn t entirely obvious, however. Reputational concerns could impact the quality of loans banks originate, even in an originate-to-distribute setting. 10

11 Section 3a: The Economics of the Securitization Structure The bankruptcy remote trust established by investment banks for the purpose of a securitization transaction has cash outflows and inflows. The largest cash outflow is the cost of purchasing the portfolio of mortgage collateral from loan originators. The largest cash inflow is the gain on the sale of the bonds issued by the trust and sold to investors. Because of this, the credit rating of a securitization deal impacts the cash inflows of the trust, also referred to as the cost of funds. Bonds rated AAA and investment grade have lower coupon payments (higher prices) then poorly rated bonds. 14 Thus, the gains of a securitization deal should be increasing in the overall credit rating of the deal. This is the basis for our assumption that banks will seek to purchase loans that maximize the potential credit rating of a deal. Figure 2, which presents an example deal structure using an actual deal in our sample, documents a monotonic increase in coupon payments as tranche credit quality decreases. This pattern should be consistent across all the deals in our sample, as coupon payments should be positively related to tranche credit quality. Section 3b: Credit Ratings, Default Correlation, and Housing Market Appreciation In this section we explain how rates of house price appreciation and collateral diversification could impact the economics of a securitization deal. We begin with a discussion of the ratings process. In a cross section of subprime securitization deals, the proportion of a deal s principal assigned a AAA rating depends on two factors; the expected loss on the pool of mortgage collateral and the correlation of default in that collateral. 15 Credit ratings reflect a rating agencies assessment of the likelihood of bond default. As an example, assume that a AAA rating assigned by a rating agency to a bond corresponds to a 1% probability that the bond will default. Default on a AAA bond occurs when the entire principal that is junior to the AAA tranche is eroded on account of loan defaults. Thus, a bond with a AAA credit rating reflects an opinion of the agency 14 The term tranches can essentially be used interchangeably with the word bonds in this setting. 15 Technically, Fitch and Standard and Poor s estimate the probability of default. Moody s estimates expected loss, which is the probability of default multiplied by the expected loss conditional on default. In practice, this difference in ratings methodology does not appear to create any substantial differences in original credit ratings. In our sample of rated tranches, in cases where S&P and Moody s have both rated a tranche, there is very rarely a disagreement in the rating and the discrepancies are minor when they exist. 11

12 that there exists a 1% probability that all the collateral junior to a senior tranche will be eroded within the expected life of the AAA bond. In order to determine the probability of default, a rating agency must estimate expected loss on the collateral pool. The expected loss in a pool of collateral is estimated as a function of loan attributes such as FICO scores, loan-to-value ratios, mortgage type, income documentation, loan purpose (refinance vs. purchase), and macroeconomic conditions, including house prices. The impact of specific loan attributes on loan default rates is documented by Sherlund (2008), Deng, Quigley, and Van Order (2000), and Pennington-Cross and Ho (2006). Loans with high FICO scores, low loan-to-value ratios, and low debt-to-income ratios default less frequently. It also has been shown that rates of house price appreciation have a strong negative association with default rates (Gerardi, Shapiro, and Willen (2007) and Sherlund (2008)). 16 While the expected loss matters for the entire deal, the shape of the expected loss distribution impacts the amount of the deal principal that can be rated AAA. The shape of the loss distribution depends on the default correlation in the underlying collateral. Default correlation measures the extent to which defaults will occur at the same time. If loan defaults are correlated, the probability of experiencing a greater percentage loss is higher, even though the expected loss remains the same. That is, default correlation simply shifts the shape of the loss distribution. Default correlation also matters in the pricing of CDOs. Longstaff and Rajan (2008) demonstrate that the expected clustering of corporate defaults explains 27% of the CDX spread. Cowan and Cowan (2004) document the degree of default correlation in a pool of subprime loans for one lender and find that the magnitude of default correlation increases as an internally assigned risk grade declines. The importance of default correlation in the context of ratings is best understood by example. Figure 3 serves as a helpful graphical reference for the following argument. First assume a scenario where the collateral has zero default correlation. If a deal has 80% of the principal rated AAA, the rating agency is estimating that the probability the 16 Though house price appreciation may not directly cause default, as house prices stagnate or decline, some homeowners are left with little or no equity and have little incentive to continue making mortgage payments. In the case of adjustable-rate mortgages, stagnant house prices prevent homeowners facing a payment reset from refinancing if the price decline has left the homeowner owing more on the mortgage than the current market value of the home. 12

13 20% of principal junior to the AAA tranche will be eroded due to defaults is 1%. Now assume a pool exists where collateral default is highly correlated. When collateral default is highly correlated, there exists a much larger probability that the 20% of collateral junior to the senior tranche will default. On the flip side, if defaults are correlated, there also exists a probability that none of the collateral will default. Thus, when the default correlation is high, the structure requires more protection for the senior tranche. In this way, for a given expected loss distribution, default correlation affects the size of the AAA tranche. We compute two empirical proxies of default correlation in our empirical tests. The first is a measure of geographic concentration of the mortgage collateral, and the second measures the covariance of housing returns in the mortgage collateral. The preceding discussion addressing expected loss and default correlation outlines the basis of our main hypothesis regarding the impact of house price appreciation. In a cross section of deals, if a rating agency determines that certain pools of loans are likely to benefit from high rates of house price appreciation relative to other mortgage pools, all else equal (including default correlation), the expected loss will be lower on the pool with high rates of house price appreciation, and less subordination will be required of those deals. The cost of funding the deal will also be lower. To the extent that mortgage pools concentrated in areas with high rates of house price appreciation receive more favorable credit ratings (again, all else equal), deal arrangers could rationally purchase mortgages of a lower marginal credit quality that are concentrated in areas with high rates of high price appreciation and still obtain the investment-grade ratings required to profitably market a securitization deal. Empirically, we test the implications of this theory in three ways. First, we test whether deals with loans concentrated in areas with high rates of expected house price appreciation are indeed able to get a larger portion of the securitization rated AAA and investment grade, controlling for default correlation. We then test whether house price appreciation impacts the cost of funds for a deal. Second, we test whether this impacts the types of loans that investment banks targeted for the purpose of securitization. We examine whether pools of loans concentrated in areas with high rates of house price appreciation are of a lower average credit quality. Finally, a potential consequence of purchasing loans of marginal credit quality masked by high house price appreciation is 13

14 the potential for higher default rates when rates of house price appreciation slow down. As such, we analyze the empirical association among credit quality, realized rates of house price appreciation, and deal-level default rates in our sample of subprime securitization deals. Section 4: Data and Summary Statistics Our empirical work relies primarily on the intersection of two data sets. The first provides detailed information on individual subprime mortgage loans. LoanPerformance, a subsidiary of First American Trust, reports information on borrower attributes and loan types for about 75% of all subprime mortgage loans originated over the past 10 years. The second set of data contains summary information on the structure and rating of residential mortgage-backed securitization deals and comes from ABSNet, a subsidiary of Standard and Poor's. The deal summary from ABSNet contains data on the total size of the securitization deal as well as the size and original credit rating of each tranche included in the deal. We measure house price appreciation for the ZIP code, Metropolitan Statistical Area (MSA), or state of each individual loan using house price data from Fiserv Case Shiller Weiss and the Office of Federal Housing Enterprise Oversight (OFHEO) in the case of state-level house prices. We obtain state-level unemployment data made available by the Census. The data appendix contains a more detailed description of each of the sources used to obtain our final dataset. Our primary unit of analysis is at the deal level. We take the following steps to identify and aggregate residential subprime securitization data to the deal level. First we obtain the deal summary for residential mortgage-backed securitization deals originated between 1997 and 2007 from ABSNet. The deal summary from ABSNet includes information on the date of issuance and the total deal amount. It also includes the original credit rating, original principal amount of each tranche, and tranche CUSIPs (each tranche, or bond has a unique CUSIP). ABSNet does not classify the residential securitization deals as being subprime. We rely on the classification of subprime deals provided by LoanPerformance. No unique numerical identifier exists between the deal summary data from ABSNet and the LoanPerformance database, so we match by hand using deal names. The total number of subprime deals included in our sample is dictated 14

15 by the number of subprime deals in the LoanPerformance database that can be matched to the universe of ABSNet deals by hand, which totals 1,315 subprime deals. We drop 64 deals from the sample because they do not have sufficient ratings information to properly quantify the structure of the deal. We double check that our hand-matching process correctly matched the LoanPerformance and ABSNet data by examining a sub-sample of deal names and deal summaries from Bloomberg. We use the deal summary from ABSNet and complementary data from Bloomberg which classifies each tranche s bond type to compute the amount of each deal rated AAA or investment grade. When possible, we use the S&P rating to determine the original credit rating of each tranche. In the few cases where S&P ratings do not exist, we use ratings provided by Moody s. The proportion of a deal rated investment grade is calculated as the sum of the N tranches with investment grade ratings divided by the total deal balance, as follows: Fraction Invest n principal AAA, AA, A, BBB+ i= 1. Grade =. deal principal To give an example, if a deal has a AAA, AA, A, and BBB+ tranche each with a $50 million balance, and the deal has a total principal balance at origination of $210 million, then 95.2% of the deal is rated investment grade. Alternatively, the subordination of the investment-grade tranches would be 4.8%. In this way, subordination measures the credit support provided to the entire investment-grade portion of the deal. Technically, subordination can be computed for each tranche. In our sample, the median deal has 16 tranches. For our purposes, the fraction of the deal rated investment grade is chosen as the relevant ratings measure because the supply of capital to securitization deals, especially from institutional investors, flows primarily to investment-grade tranches. We provide the details of an example deal from our sample in Figure 2. Figure 2 reports the amount of principal contained in each tranche, the original credit rating, the first coupon payment that was made to investors and the coupon spread over 1-month LIBOR. The median securitization deal in our sample has 5,219 mortgage loans serving as collateral. We aggregate the individual loan-level data from LoanPerformance to the deal level by taking the loan-weighted average of each deal attribute. We outline this 15

16 process in the data appendix. We match rates of house price appreciation to the data in three steps. First we match ZIP-code house price indexes to individual loans according the ZIP code reported in the loan documentation from Loan Performance. If a house price index is not available for the ZIP code, we match MSA-level house prices. 17 Finally, if an MSA index is not available for the loan, we use a state-level house price index. After matching individual loans with their respective rates of house price appreciation we aggregate the house price appreciation rates to the deal level using individual loan sizes within each deal as weights. Finally, we merge unemployment rates to the deal level using the state unemployment data. The final data set includes 1,251 securitization deals, 6.7 million loans that serve as collateral in the deals, unemployment rates from 50 states, and house price appreciation data at the ZIP code, MSA, and state levels. Section 4b: Summary statistics Table 1 reports summary statistics on the attributes of securitization deals for the entire sample, which runs from We report the mean, median, standard deviation, and extreme percentiles of key deal attributes. The average subprime deal has a loan-weighted median FICO score of 621, a median loan-to-value ratio of 84% and a debt-to-income ratio of 41%. The median proportion of the deal with investment-grade ratings is 94.5%, while the median deal has an excess spread of 361 basis points. The median unemployment rate at deal origination was 5.1%, median house price appreciation in the year preceding deal origination was 13.7%, and median overvaluation was 7%. 18 Table 2 reports summary statistics for deals originated through time and reveals a substantial increase in the number of securitization deals originated over the last decade. More deals were originated in 2005 and 2006 alone than in the entire preceding nine years combined. The principal included in the deals was also substantially larger, having increased from a median size of $478 million in 2000 to $1.02 billion in Table 2 also highlights an important trend in the structure of securitization deals through time. The proportion of each deal that was rated investment grade declined almost 17 ZIP-code level house price data is available for about a third of the loans in our sample. MSA-level data is available for a large portion of the remaining two-thirds of our sample. 18 We discuss the overvaluation concept in the following section. 16

17 monotonically since In 2001, the typical deal had 97.1% of the principal rated investment grade, while the typical deal had just 88.2% rated investment grade in The monotonic decline appears to be in response to a marked inter-temporal decline in the amount of excess spread for the typical deal. 20 Table 2 also documents the time series pattern in the default rate of loans in each deal. The default rate is calculated as the total number of loans in the process of foreclosure or already foreclosed in the year after the deal was originated divided by the total number of loans in the securitization deal. As an example, the numerator in the default rate for deals originated in 2006 is calculated as the total number of defaulted loans by the end of The jump in the default rate from 5.6% in 2005 to 13.7% in 2006 is associated with a significant decline in the average rate of house price appreciation over the same period. Table 3 reports the time series attributes of the loan characteristics of each deal as well as average rates of house price appreciation. FICO scores generally increased through the sample period, as did loan-to-value and debt-to-income ratios. 21 The percentage of loans originated with adjustable rates increased substantially through the period. In general, aside from FICO scores, Table 3 reveals deterioration in the quality of loans being securitized through time, a pattern also identified by Demyanyk and Van Hemert (2007). Rates of house price appreciation rose dramatically over the sample period, reaching their peak in The unemployment rate varied little, ranging from a low of 3.97% in 2000 to a high of 6.13% in Section 5: Empirical Methodology and Results Our empirical analysis is designed to test the benefits of diversification and the impact of differing rates of house price appreciation on the structure of securitization deals. The structure of a securitization deal should matter because it could impact the cost of funding the mortgage loans purchased for the securitization. We begin by developing proxies for collateral diversification. We then test how expected rates of house price 19 The same pattern exists for the percent of a deal rated AAA. 20 Recall that subordination and excess spread are both forms of credit support. Thus, they can be viewed as substitutes, albeit imperfect ones. 21 Debt-to-income ratios are missing for some loans early in the sample. We estimate some of our empirical tests with and without debt-to-income ratios to preserve sample size. 17

18 appreciation impact the ratings and cost of funds for a deal. Our final set of tests analyzes the implications of our hypothesis. We analyze whether rates of house price appreciation are associated with the credit quality of loans banks are willing to purchase, and analyze the association between credit quality and loan performance. Section 5a: Constructing Measures of Collateral Diversification We begin by developing two empirical proxies for collateral diversification. We measure geographic diversification by constructing a Herfindahl index, a calculation which measures the geographic concentration of the mortgage collateral. Our second proxy for diversification, referred to as housing market diversification, measures the covariance in the housing market returns in a portfolio of mortgage loans. Our motivation for constructing two separate measures of diversification is straightforward. Geographic diversification does not guarantee diversification in housing market returns. To the extent that the housing market is a primary factor in the probability of loan default, a relevant measure of loan diversification is the correlation between the returns in housing markets of the loan collateral. As an illustration of this point consider that despite geographic distance, returns on a California house price index have a correlation coefficient of 0.87 with returns on an index measuring house price returns in Washington DC. 22 We construct a Herfindahl index of the geographic concentration in each deal as follows. For each deal we calculate the percentage of the deal principal concentrated in each of the 51 states (Washington DC enters the calculation separately). The deal-level Herfindahl index is then calculated as the sum of the squared weights, expressed n 2 w i i= 1 as. We report summary statistics on deal-level Herfindahl measures of geographic diversification in Tables 1 and 3. Not reported in Table 3 is the fact that in our sample, the average deal has 28% of total loan principal concentrated in California The correlation is calculated using the state-level repeat sales house price index from OFHEO. The data are quarterly and run from See Mayer and Pence (2008) for a more thorough analysis of the geographic dispersion in subprime loan originations. 18

19 We construct deal-level measures of the diversification in housing market returns in the following way. Again, for each deal, we calculate the percent of deal principal concentrated in each of the 51 states. The total portfolio covariance in housing market 2 returns is then calculated as W '( Σ σ i )W, where W is a 51x1 matrix of loan concentration weights, and Σ is a 51x51 variance-covariance matrix of housing market returns. We subtract the variance of each housing market from the weighted variancecovariance matrix because we are only interested in the covariance of the housing market returns, not the variance of an individual market. Thus the variance-covariance matrix has zeros in the diagonal. The calculation results in a 1x1 scalar which is a summary measure of the covariance in housing market returns for each deal, with the covariance matrix weighted by the loan concentration in each state. The intuition is as follows. A deal that is highly concentrated in two states whose housing markets are historically highly correlated (not geographically) will have a larger covariance, and thus a higher probability of experiencing housing market declines at the same time. Summary statistics of market diversification are also reported in Tables 1 and 3. Section 5b: Market Fundamentals, House Price Appreciation, and Deal Structure Having constructed proxies for collateral diversification we turn our focus to expected rates of house price appreciation. Our primary hypothesis proposes the existence of a positive relationship between expected rates of house price appreciation and the proportion of the deal rated investment grade. If a pool of loans is likely to benefit from high rates of house price appreciation over the life of the deal, less subordination will be required of the deal, and the cost of funding the deal will be lower. This relationship should have some impact on the types of loans investment banks seek to purchase for the purpose of securitization. Testing this hypothesis requires a proxy for the expectation of rates of house price appreciation over some future time horizon. 24 Using actual rates of appreciation as a proxy for expected rates, a common approach in models of empirical asset pricing, is not a viable option. This is because our hypothesis suggests the existence of a relationship between deal structure in the secondary market and the eventual level of credit supplied to the primary market. To the 24 The exact time horizon will vary by deal, but most of the principal for most deals lasts 3-5 years. 19

20 extent that credit supply in the primary market influences subsequent rates of house price appreciation, the central argument of Sufi and Mian (2008), realized rates of house price appreciation measured over a given time period after deal origination may be endogenously determined by the deal structure itself. In our attempt to overcome this endogeneity we use two proxies for expected rates of house price appreciation at the time of deal origination. Our first proxy for expected house price appreciation relies on housing market fundamentals to construct a measure of whether a housing market is expensive or cheap relative to its long run equilibrium. Theoretically, home prices should demonstrate an equilibrium relationship with rents and the user costs of housing. As discussed in Gallin (2004) and Himmelberg, Mayer, and Sinai (2005), the fundamental relationship between house prices and rents is commonly expressed as: i t R t p y Pt [( it + τ t )(1 τ t ) + δ t + λt EtG + 1], = t τ t τ t P Y where is the real interest rate, and represent property and marginal income tax rates, δ t is the maintenance and depreciation rate, λt is a housing market risk premium, and EG t + 1 represents expected capital gains. The expression + τ p y ( i )( 1 τ ) δ + λ E G t t t + t t t t+ 1 is referred to as the user cost of owning a home and represents the cost and benefit of taxes, depreciation and maintenance, expected capital gains, and the financial risk of owning a property. Multiplying the user cost of housing by the level of house prices generates an imputed rent. The fundamental relationship between imputed rents and actual rents compares the true cost of owning a property for a year against the cost of renting an equivalent property. As stated succinctly by Himmelberg, Mayer, and Sinai (2005), [I]f annual ownership costs rise without a commensurate increase in rents, house prices must fall to convince potential homeowners to buy instead of renting. The converse happens if annual ownership costs fall. Our measure of expected price appreciation relies on the imputed rent to actual rent ratios first constructed by Himmelberg, Mayer, and Sinai (2005) and since updated 20